4 * kernel/hung_task.c - kernel thread for detecting tasks stuck in D state
10 #include <linux/nmi.h>
11 #include <linux/init.h>
12 #include <linux/delay.h>
13 #include <linux/freezer.h>
14 #include <linux/kthread.h>
15 #include <linux/lockdep.h>
16 #include <linux/export.h>
17 #include <linux/sysctl.h>
18 #include <linux/suspend.h>
19 #include <linux/utsname.h>
20 #include <linux/sched/signal.h>
21 #include <linux/sched/debug.h>
23 #include <trace/events/sched.h>
26 * The number of tasks checked:
28 int __read_mostly sysctl_hung_task_check_count = PID_MAX_LIMIT;
31 * Limit number of tasks checked in a batch.
33 * This value controls the preemptibility of khungtaskd since preemption
34 * is disabled during the critical section. It also controls the size of
35 * the RCU grace period. So it needs to be upper-bound.
37 #define HUNG_TASK_BATCHING 1024
40 * Zero means infinite timeout - no checking done:
42 unsigned long __read_mostly sysctl_hung_task_timeout_secs = CONFIG_DEFAULT_HUNG_TASK_TIMEOUT;
45 * Zero (default value) means use sysctl_hung_task_timeout_secs:
47 unsigned long __read_mostly sysctl_hung_task_check_interval_secs;
49 int __read_mostly sysctl_hung_task_warnings = 10;
51 static int __read_mostly did_panic;
52 static bool hung_task_show_lock;
53 static bool hung_task_call_panic;
55 static struct task_struct *watchdog_task;
58 * Should we panic (and reboot, if panic_timeout= is set) when a
59 * hung task is detected:
61 unsigned int __read_mostly sysctl_hung_task_panic =
62 CONFIG_BOOTPARAM_HUNG_TASK_PANIC_VALUE;
64 static int __init hung_task_panic_setup(char *str)
66 int rc = kstrtouint(str, 0, &sysctl_hung_task_panic);
72 __setup("hung_task_panic=", hung_task_panic_setup);
75 hung_task_panic(struct notifier_block *this, unsigned long event, void *ptr)
82 static struct notifier_block panic_block = {
83 .notifier_call = hung_task_panic,
86 static void check_hung_task(struct task_struct *t, unsigned long timeout)
88 unsigned long switch_count = t->nvcsw + t->nivcsw;
91 * Ensure the task is not frozen.
92 * Also, skip vfork and any other user process that freezer should skip.
94 if (unlikely(t->flags & (PF_FROZEN | PF_FREEZER_SKIP)))
98 * When a freshly created task is scheduled once, changes its state to
99 * TASK_UNINTERRUPTIBLE without having ever been switched out once, it
102 if (unlikely(!switch_count))
105 if (switch_count != t->last_switch_count) {
106 t->last_switch_count = switch_count;
107 t->last_switch_time = jiffies;
110 if (time_is_after_jiffies(t->last_switch_time + timeout * HZ))
113 trace_sched_process_hang(t);
115 if (!sysctl_hung_task_warnings && !sysctl_hung_task_panic)
119 * Ok, the task did not get scheduled for more than 2 minutes,
122 if (sysctl_hung_task_warnings) {
123 if (sysctl_hung_task_warnings > 0)
124 sysctl_hung_task_warnings--;
125 pr_err("INFO: task %s:%d blocked for more than %ld seconds.\n",
126 t->comm, t->pid, timeout);
127 pr_err(" %s %s %.*s\n",
128 print_tainted(), init_utsname()->release,
129 (int)strcspn(init_utsname()->version, " "),
130 init_utsname()->version);
131 pr_err("\"echo 0 > /proc/sys/kernel/hung_task_timeout_secs\""
132 " disables this message.\n");
134 hung_task_show_lock = true;
137 touch_nmi_watchdog();
139 if (sysctl_hung_task_panic) {
140 hung_task_show_lock = true;
141 hung_task_call_panic = true;
146 * To avoid extending the RCU grace period for an unbounded amount of time,
147 * periodically exit the critical section and enter a new one.
149 * For preemptible RCU it is sufficient to call rcu_read_unlock in order
150 * to exit the grace period. For classic RCU, a reschedule is required.
152 static bool rcu_lock_break(struct task_struct *g, struct task_struct *t)
161 can_cont = pid_alive(g) && pid_alive(t);
169 * Check whether a TASK_UNINTERRUPTIBLE does not get woken up for
170 * a really long time (120 seconds). If that happens, print out
173 static void check_hung_uninterruptible_tasks(unsigned long timeout)
175 int max_count = sysctl_hung_task_check_count;
176 int batch_count = HUNG_TASK_BATCHING;
177 struct task_struct *g, *t;
180 * If the system crashed already then all bets are off,
181 * do not report extra hung tasks:
183 if (test_taint(TAINT_DIE) || did_panic)
186 hung_task_show_lock = false;
188 for_each_process_thread(g, t) {
191 if (!--batch_count) {
192 batch_count = HUNG_TASK_BATCHING;
193 if (!rcu_lock_break(g, t))
196 /* use "==" to skip the TASK_KILLABLE tasks waiting on NFS */
197 if (t->state == TASK_UNINTERRUPTIBLE)
198 check_hung_task(t, timeout);
202 if (hung_task_show_lock)
203 debug_show_all_locks();
204 if (hung_task_call_panic) {
205 trigger_all_cpu_backtrace();
206 panic("hung_task: blocked tasks");
210 static long hung_timeout_jiffies(unsigned long last_checked,
211 unsigned long timeout)
213 /* timeout of 0 will disable the watchdog */
214 return timeout ? last_checked - jiffies + timeout * HZ :
215 MAX_SCHEDULE_TIMEOUT;
219 * Process updating of timeout sysctl
221 int proc_dohung_task_timeout_secs(struct ctl_table *table, int write,
223 size_t *lenp, loff_t *ppos)
227 ret = proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
232 wake_up_process(watchdog_task);
238 static atomic_t reset_hung_task = ATOMIC_INIT(0);
240 void reset_hung_task_detector(void)
242 atomic_set(&reset_hung_task, 1);
244 EXPORT_SYMBOL_GPL(reset_hung_task_detector);
246 static bool hung_detector_suspended;
248 static int hungtask_pm_notify(struct notifier_block *self,
249 unsigned long action, void *hcpu)
252 case PM_SUSPEND_PREPARE:
253 case PM_HIBERNATION_PREPARE:
254 case PM_RESTORE_PREPARE:
255 hung_detector_suspended = true;
257 case PM_POST_SUSPEND:
258 case PM_POST_HIBERNATION:
259 case PM_POST_RESTORE:
260 hung_detector_suspended = false;
269 * kthread which checks for tasks stuck in D state
271 static int watchdog(void *dummy)
273 unsigned long hung_last_checked = jiffies;
275 set_user_nice(current, 0);
278 unsigned long timeout = sysctl_hung_task_timeout_secs;
279 unsigned long interval = sysctl_hung_task_check_interval_secs;
284 interval = min_t(unsigned long, interval, timeout);
285 t = hung_timeout_jiffies(hung_last_checked, interval);
287 if (!atomic_xchg(&reset_hung_task, 0) &&
288 !hung_detector_suspended)
289 check_hung_uninterruptible_tasks(timeout);
290 hung_last_checked = jiffies;
293 schedule_timeout_interruptible(t);
299 static int __init hung_task_init(void)
301 atomic_notifier_chain_register(&panic_notifier_list, &panic_block);
303 /* Disable hung task detector on suspend */
304 pm_notifier(hungtask_pm_notify, 0);
306 watchdog_task = kthread_run(watchdog, NULL, "khungtaskd");
310 subsys_initcall(hung_task_init);